feat: add LXD emergency cleanup tool

[josecelano]

Summary

Adds an emergency cleanup tool for orphaned LXD environments that remain after interrupted test workflows or deployment failures.

Changes

• New Binary: lxd_cleanup tool for emergency cleanup of orphaned LXD resources
  • Supports dry-run mode for safe preview
  • Handles multiple environments in batch
  • Clear warnings about emergency-only usage
• Documentation: Comprehensive guide in docs/tools/lxd-cleanup.md
  • Emphasizes normal destroy + purge workflow
  • Provides troubleshooting and verification steps
• Agent Skill: Added clean-lxd-environments skill in .github/skills/
  • On-demand workflow activation for cleanup scenarios
  • Includes verification and troubleshooting guidance

Motivation

During E2E testing and development, LXD VMs can become orphaned when:

• Tests are interrupted (Ctrl+C)
• Deployment fails mid-process
• State desynchronizes between application and LXD

The normal destroy + purge workflow requires valid state, making it unsuitable for these scenarios.

Usage

Emergency only - prefer normal workflow:

cargo run -- destroy environment-name
cargo run -- purge environment-name

When normal workflow fails:

# Preview cleanup
cargo run --bin lxd_cleanup -- --dry-run environment-name

# Perform cleanup
cargo run --bin lxd_cleanup -- environment-name

Testing

• All linters passing (markdown, YAML, TOML, cspell, clippy, rustfmt, shellcheck)
• Validated against 8 orphaned VMs from previous testing
• All skill commands tested and working

References

• Related: E2E testing cleanup patterns
• Documentation: docs/tools/lxd-cleanup.md
• Skill: .github/skills/clean-lxd-environments/skill.md

[josecelano]

ACK 0e279c0